Analysis of Inter-observer Variation From Meningioma eContouring Session

Abstract

Inter-observer contour reproducibility may represent a surrogate measure of consensus in delineation of the region at risk from the disease process. We sought to quantify the inter-observer variation in delineating both target and organs-at-risk (OAR) for patients with atypical meningioma. Permission was obtained from the American Society of Radiation Oncology (ASTRO) to analyze inter-observer variation in physician attendee contours of atypical meningioma at the 2017 ASTRO eContouring session. Twenty-nine radiation oncologists participated. They were provided a case history of partially resected atypical meningioma abutting the left optic nerve. Pre- and post- operative MR imaging with diagnostic reports were available. Each participant contoured the gross tumor volume (GTV), clinical target volume (CTV), brainstem, optic chiasm, and optic nerves. The data sets were anonymized. Dice statistics to assess the degree of overlap between an individual’s contour against a referenced expert, descriptive statistics, and coefficients of variation (SD/mean) were calculated for the GTV, CTV, and OARs. Zero overlap would yield a Dice coefficient of 0.00, while perfect overlap a coefficient of 1.00. Similar to other studies where Dice coefficients of 0.85-0.90 indicate consistency between observers, we set a Dice coefficient of <0.80 to indicate meaningful variation. The highest mean Dice coefficient and lowest coefficient of variation was observed for the brainstem. While preoperative GTV demonstrated a high degree of agreement between observers, there was poor agreement in postoperative planning GTV and CTV (Table 1). The difference between GTV and CTV mean Dice statistics was not statistically significant (p=0.09). Poorest agreement was observed for the target-approximate optic chiasm. Compared to other primary diseases treated with radiotherapy, there is considerable inter-observer variability in GTV and CTV contours of atypical meningioma. This may represent both inconsistent training for meningioma treatment planning and a deeper lack of agreement regarding delineation of gross and subclinical disease. This variability may result in both under-treatment and OAR overexposure. This has particularly serious implications for meningioma radiosurgery, which is increasingly used in spite of a paucity of long-term follow-up. Evidence-driven guidelines defining optimal target delineation of meningiomas are warranted.Abstract 2144; Table 1StructureDice (mean)SDCoefficient of Variation (SD/mean)Brainstem0.8890.0720.081Chiasm0.4830.2000.414CTV0.6880.1160.169GTV0.7460.1270.170LON0.5900.1620.275Pre-sx tumor0.9100.0140.016RON0.6070.1230.202StructureVolume (median cc)MinMaxGTV5.0900.72018.040 Open table in a new tab